Type | |
Stats | 503 |
Reviews | (81) |
Published | Feb 9, 2024 |
Base Model | |
Hash | AutoV2 7CA634404D |
V2.0
Telegram: https://t.me/+nbU3j7rLEYZkNzVl
Tutu held in his Q group: Zhao Tutu's Visual Miracle Challenge, this is a long-term activity, as long as you share high-quality drawing works in the group, you may get a unique community title + major modeling sites VIP and arithmetic + Tutu's private modeling + physical rewards + mysterious gifts + omni-channel publicity + communication and Q&A, come and play with me! Exchange Q group: 950351015;
Epic Enhancement.
Made after many rounds of training and fusion.
The main purpose is to optimize the face as well as improve the rate of good pictures, basically the purpose has been achieved.
Rigorous data cleaning, scientific training and fusion were carried out. All kinds of 2.5D maps can be easily held.
Suggestions for the map:
VAE: no need
Iteration steps: 80 or less, I often use 40
Sampling method (Sampler): DPM + + 2M SDEKarras, other will do.
High-resolution fix (Hires.fix): I often use 4x-UltraSharp, others are fine.
High Score Iteration Steps: I often use 20 Redraw Amplitude: this is determined by the needs of the outgoing map.
Magnification: 2
Width: 540 (512) Height: 960 (768)
Cue word guidance factor: 7
ADetailer: on (far on, close you think not on is fine) face_yolov8n_v2.pt ;)
图图在自己的Q群里举办了：赵图图的视觉奇迹挑战活动，这是一个长期活动，只要在群里分享优质绘图作品，就有可能获得独特社区称号+各大模型网站VIP及算力+图图的私模+实物奖励+神秘礼品+全渠道宣传+交流答疑，快来和我一起玩吧！交流Q群：950351015；
史诗级增强。
经过多轮训练和融合而成。
主要的目的是优化面部以及提高良图率，基本上目的已经达到。
进行了严谨的数据清洗，科学的训练与融合。 各种2.5D图都能轻松拿捏。
出图建议：
VAE：不用
迭代步数：80以内，我经常用40
采样方法 (Sampler)：DPM++ 2M SDEKarras，别的也行。
高分辨率修复 (Hires.fix)：我经常用4x-UltraSharp，别的也行。
高分迭代步数：我经常用20 重绘幅度：这个要根据出图的需求确定。
放大倍数：2
宽度：540（512） 高度：960（768）
提示词引导系数：7
ADetailer：开启（远的开，近的你觉得不开也行）face_yolov8n_v2.pt ;
V1.0
I don't know why I suddenly wanted to make a big model, so I did.
Still used a lot of stylised real photos and fine-tuned the training out.
Used some of my own photoshopped in there and adjusted the style appropriately.
Usage:
1, VAE is already included. Of course, if you want to hang out individual what can also be.
2, suggested starting: high-resolution image, natural light, Asian ethnicity,young woman, and then write 1girl. suggested directly with Asian ethnicity,young woman instead of 1girl, the effect is more obvious. Of course, this is out of the case of Asian young woman.
Hair can be written long hair, bangs, because I myself is this hairstyle ...
3, the number of iterative steps I often write 50, according to their own habits combined with other parameters to determine it.
4, sampling method: I often use DPM + + 2M SDE Karras, of course, other can be.
5, high-resolution repair: zoom algorithm: 4x-UltraSharp, high iteration step I often use 40, repainting amplitude here to note that if you macro, or dislike the face of the map, high repair must not be adjusted too high, 0.2 to 0.35 between enough, if it is a whole body or panorama, you can be appropriately pulled up a little bit, and open the face repair (ADetailer)!
6, width and height, I commonly used 540 960, of course, 512 768 is also a good choice.
7, CFG: I commonly used 7.
8, ADetailer only out of the whole body or panoramic view of the time to open. Don't open it when you are in close proximity.
Note that these parameters are not necessarily the best parameters, according to their own needs to try more ha.
If you encounter problems in use, you can leave me a message.
Looking forward to your work!
Looking forward to your work!
Looking forward to your work!
Statement:
1, You should take full responsibility for any creative works using this model.
2、You must not use this model to intentionally create or share illegal or harmful output or content, and avoid using this model for malicious, hurtful, defamatory, fraudulent or political purposes.
3、Please notify me if you use this model for commercial purposes, thank you.
Frequently Asked Questions:
1、Why can't I produce the same photos as yours?
Everyone's computer configuration and operating environment are different, the use of LORA and plug-in version, plug-in configuration, etc. will also make a difference, so please do not pursue the exact same photo, create your own perfect work!
2、How to get in touch?
If there is a demand for custom models, or other cooperation intentions, you can directly leave a message here, or contact Q331506796 (specify the intention)
Hope you can give this model a chance, I believe it will not let you down!
不知道为什么忽然就想做个大模型了，所以就做了。
还是用了很多风格化的真实照片，微调训练出来。
用了一些自己的照骗在里面，并对风格进行了适当调整。
使用方法：
1、VAE已经包含。当然如果你要外挂个别的什么也可以。
2、建议起手： high-resolution image, natural light, Asian ethnicity,young woman，然后再写1girl。建议直接用Asian ethnicity,young woman代替1girl,效果更明显。当然，这是出亚洲小姐姐的情况时的情况。
发型可以写long hair, bangs, 因为我自己是这个发型...
3、迭代步数我很多时候写的50，根据自己的习惯综合其他参数判断吧。
4、采样方法：我经常用DPM++ 2M SDE Karras,当然别的也是可以的。
5、高分辨率修复：放大算法：4x-UltraSharp,高分迭代步数我常用40，重绘幅度这里要注意，如果你微距，或者怼着脸出图，高分修复一定不要调的太高，0.2到0.35之间就够了，如果是出全身或者全景，就可以适当拉大一点点，并且打开修脸（ADetailer）
6、宽高，我常用540*960，当然512*768也是一个不错的选择。
7、CFG:我常用7。
8、ADetailer只有出全身或者全景的时候再开。近距离的时候不要开。
注意，这些参数并不一定是最佳参数，要根据自己的需要多多尝试哈。
如果您在使用中遇到问题，可以给我留言。
期待您的作品！
期待您的作品！
期待您的作品！
声明：
1、您应对使用此模型的任何创意作品承担全部责任。
2、您不能使用该模型故意制作或分享非法或有害的输出或内容，避免将此模型用于恶意、伤害、诽谤、诈骗或政治用途。
3、如果将本模型用于商业用途，请通知我，谢谢。
常见问题：
1、为什么我出不了和你一样的照片？
每个人的电脑配置以及操作环境都不一样，使用的LORA以及插件版本、插件配置等也会有所区别，所以请不要追求一模一样的照片，创作您自己的完美作品吧！
2、如何取得联系？
如果有定制模型的需求，或者有其他合作意向，可以直接在这里留言，或联系Q331506796（注明来意）
希望可以给这个模型一个机会，相信不会让你失望的！